University of North Texas
A Hybrid SQL and Retrieval-Augmented Generation System for Natural Language Querying of Friction Stir Additive Manufacturing Research Data
Abstract
dc:descriptionThe growing volume of Friction Stir Additive Manufacturing (FSAM) research literature presents a significant challenge for researchers seeking to efficiently retrieve and compare experi- mental findings across publications. Existing approaches require either programming knowledge for database queries or time-consuming manual literature search. This thesis presents the design, implementation, and evaluation of the FSAM Research Assistant, a hybrid natural language query system that combines a structured SQL database of 57 peer-reviewed FSAM papers with a retrieval- augmented generation pipeline over a vector database of 1103 semantically indexed text chunks. The system employs a smart router to classify user questions into SQL, retrieval-augmented gen- eration, or uploaded paper search paths. The SQL pipeline uses fuzzy alloy name matching and large language model based natural language to SQL generation. The retrieval-augmented gener- ation pipeline uses BAAI/bge-large-en-v1.5 dense embeddings and ChromaDB similarity search, with LLaMA 3.3 70B via the Groq API for answer generation. A Streamlit frontend and FastAPI backend provide a fully deployable research interface accessible without dedicated GPU infrastruc- ture on standard university hardware. Manual evaluation confirms that the SQL pipeline reliably retrieves correct structured data, while the retrieval-augmented generation pipeline provides use- ful explanatory responses for broad mechanism questions with observed inconsistency for narrow domain-specific queries.
Degree
thesis:*- Grantor dc:publisher
- University of North Texas
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bhagat, Yashvikumari Niravkumar 2001-
- Contributors dc:contributor
-
- Choudhary, Sagnik Ray
- Bhowmick, Sanjukta
- Mishra, Rajiv S.
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Public
- Bhagat, Yashvikumari Niravkumar 2001-
- Copyright
- Copyright is held by the author, unless otherwise noted. All rights Reserved.
- Language dc:language
- English
Identifiers
dc:identifier.*- Identifier
-
local-cont-no: submission_4926
ark: ark:/67531/metadc2652436